Langfuse bets open-source LLM observability wins; ClickHouse acquires it

Open-source LLM engineering platform: tracing, prompt management, evals, datasets and a playground for teams building and debugging LLM applications, run self-hosted or as managed cloud.

That teams shipping LLM apps to production would adopt an open, self-hostable tracing/eval layer, and open source would make it the default before closed SaaS could.

$4M seed announced November 7, 2023, led by Lightspeed Venture Partners with La Famiglia (General Catalyst) and Y Combinator.

Founded in Berlin by Clemens Rawert, Max Deichmann and Marc Klingen; the team entered YC W23 with a SaaS billing prototype, pivoted three weeks in, and found Langfuse while shipping LLM prototypes that were easy to demo and hard to run. The first version was open-sourced in August 2023 and won Product of the Day on Product Hunt the same month; the $4M seed followed in November 2023.

Adoption outgrew the original architecture: Langfuse v3 replaced the Postgres core with ClickHouse to handle high-throughput ingestion and fast analytical reads, and the team opened a second office in San Francisco. By the end of 2025 the open-source project had passed 20k GitHub stars with 26M+ SDK installs a month, and Langfuse said it was trusted by 19 of the Fortune 50 and 63 of the Fortune 500.

Acquired by ClickHouse in January 2026. The founders had term sheets for a Series A, but chose the acquisition instead: the whole team joined ClickHouse, Langfuse stayed MIT-licensed and self-hostable, and ClickHouse folded LLM observability into its platform alongside a $400M Series D announced the same day.

Every LLM team hit the same production problem — demos work, debugging doesn't — so the founders sold a tool to a pain they had themselves, not a market they guessed at.

Open source and self-hosting let Langfuse spread inside enterprises without a sales team, turning engineers into distribution and removing vendor-lock-in objections.

MIT licensing and framework-agnostic SDKs made it safe to adopt next to LangChain, LlamaIndex, OpenAI and LiteLLM, so it became the neutral layer of the stack.

By the time of acquisition Langfuse was a dependency thousands of teams ran — 20k stars and 26M monthly SDK installs — so ClickHouse bought the feedback loop instead of building it.

Choosing acquisition over a Series A traded near-term independence for ClickHouse's engineering and enterprise reach, letting the founders keep building open source with more resources.

Open source made Langfuse default infrastructure: give away the tracing layer every LLM team needs, and the tool thousands run becomes infrastructure another vendor must buy.

As of September 2026, Langfuse operates inside ClickHouse: the MIT-licensed core, self-hosted deployments and Langfuse Cloud are unchanged, and the Berlin/San Francisco team continues shipping, including Claude Code integration and a v4 beta (per the GitHub repo and Langfuse's January 2026 update). ClickHouse markets it as its native LLM observability layer. The open question is whether staying independent would have built a bigger standalone business — the founders bet that joining ClickHouse made it faster instead.